1
$\begingroup$

I am trying to decide what GPU or GPUs to buy to run tf-pose pose detection and yolo3 object detection on several cameras. I need to keep an acceptable frame rate too.

what kind of GPU configuration should I go for?

does having multiple GPUs have an effect on what I am trying to do.

can you please explain the difference between using 2-4 1080ti(any GPU in that price range) cards and using something 10 times more expensive than that such as v100.in training and using the networks both

$\endgroup$
1
  • 1
    $\begingroup$ Hello. Asking for recommendations (about hardware) is generally off-topic here. This type of question may be more suited for Data Science SE (but I don't know if it's really on-topic there). Please, take a look at ai.stackexchange.com/help/on-topic to know more about our scope. $\endgroup$ – nbro Sep 29 '20 at 22:33
1
$\begingroup$

Here's a link to some benchmarks that should give you some insights.

In my experience (I've used systems with both 1080s and V100s) I've found that as of about a year ago, a lot of the common tools couldn't use the V100s well. Until we started doing some manual optimization, the 1080s were comparable if not better on common tasks. Of course, once we put in the time to optimize, recompile drivers, etc., the V100's really came to life and outperformed everything else by several times.

$\endgroup$
5
  • $\begingroup$ how about the training and predicting part $\endgroup$ – Eshaka Jan 19 '20 at 4:47
  • $\begingroup$ What about it? Comparable performance for common tasks and tools as long as you're keeping your batches small enough to fit into memory for both cards. If memory serves, for YOLO, a 1080 can process a 24 frames/second video feed in realtime, I believe the V100 could do simultaneous feeds with some optimization. $\endgroup$ – user1209675 Jan 20 '20 at 15:52
  • $\begingroup$ so memory is not important if processing one frame at a time? $\endgroup$ – Eshaka Jan 21 '20 at 2:14
  • $\begingroup$ Processing one frame at a time pretty much only requires enough memory for the model, the frame data, and some overhead. It would have to be a pretty large image and model to stress something like a 1080. I think you'll be fine unless you're doing something crazy like trying to real-time process 4k video or something. $\endgroup$ – user1209675 Jan 22 '20 at 13:55
  • $\begingroup$ i do want to process large images, one at a time, maybe even 4k, but the most important thing is the speed, I most definitely need realtime good fps. is the memory is not important, what is. how to choose the best hardware $\endgroup$ – Eshaka Jan 23 '20 at 1:09

Not the answer you're looking for? Browse other questions tagged or ask your own question.